|
Activity Number:
|
513
- Bayesian Learning for the Health Care and Disparities in the 21st Century
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Thursday, August 11, 2022 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
International Society for Bayesian Analysis (ISBA)
|
|
Abstract #322309
|
|
|
Title:
|
Bayesian Nonlinear Models for Data Integration and Distributed Statistical Inference
|
|
Author(s):
|
sounak chakraborty* and Tanujit Dey and Anjishnu Banerjee
|
|
Companies:
|
university of missouri and Harvard University and Medical College of Wisconsin
|
|
Keywords:
|
RKHS;
Data Integration;
Surrogate Likelihood;
Bayesian Inference;
High dimensional
|
|
Abstract:
|
In this paper we introduce Bayesian non-linear models for combining information from multiple data source platforms brought together on the same patient set. The proposed model can efficiently combine and borrow information across platforms and can provide a detailed complementary view of a specific disease. Our model can explore high dimensional covariate space with the help of reproducing kernel Hilbert space based low dimensional representation of non-linear functions based on random Fourier transformation approximation and modified Normal-Exponential-Gamma prior. In this paper we also develop an alternative approach where we can do distributed statistical inference across multiple similar data sets under Surrogate Likelihood framework. This is useful where large volume of data is accessed from multiple data sites and information of each site need to be handled and processed locally for privacy and data security issues. Our approach relies on efficient surrogate likelihood for the global likelihood and computes a quasi-posterior distribution. All Bayesian inference is based on this quasi posterior distribution improving the computational efficiency of our MCMC algorithms.
|
Authors who are presenting talks have a * after their name.
